Claude Fable 5.2: New Model Cuts Coding Costs 20% for Students
Anthropic launched Claude Fable 5.2 today, and it's the cheapest top-tier coding model yet at $8 in and $40 out per million tokens. Here's what that means for a student budget, plus four projects you can build free during launch week.
You have ₹500 left this month for AI tools. Yesterday that bought you almost nothing. Today it buys you a weekend of coding help. Anthropic launched Claude Fable 5.2 today, 9 November 2026, and it is the cheapest top-tier coding model yet: $8 per million input tokens, $40 per million output tokens. That is 20% cheaper than Fable 5.1 and now under GPT-6 Astra, per Anthropic. Suppose you are a student working on a small coding assignment; you might send a few thousand input tokens and receive a few hundred output tokens, which would cost well under a cent. The real cost appears when you run long agent sessions that generate many output tokens.
Claude Fable 5.2 is Anthropic's newest coding model, launched 9 November 2026, priced at $8 in / $40 out per 1M tokens (about ₹670 / ₹3,350 at current rates). Students get it free on Pro during launch week until 16 November, or included in Max and Ultra plans.
What Fable 5.2 costs per token
Anthropic announced the pricing on its Fable and Mythos 5.2 page, and TechCrunch confirmed the price drop in its coverage of the launch.
| Model | Input per 1M tokens | Output per 1M tokens | Notes |
|---|---|---|---|
| Fable 5.1 | $10 | $50 | Previous generation |
| Fable 5.2 | $8 | $40 | 20% cheaper, launched today |
| GPT-6 Astra | — | — | Now above Fable 5.2's price |
No source lists a rupee price, so here is the arithmetic on the dollar figure: at current rates, $8 is about ₹670 and $40 is about ₹3,350. If you are budgeting in rupees, treat those as rough conversions of the sourced dollar prices, not as official Anthropic pricing.
What got faster and better
Fable 5.2 produces outputs up to 30% faster than Fable 5.1, with better coding, reasoning, and vision, and an optional fast mode that runs up to 2.5x faster at double the price. Anthropic's launch page lists the speed increase and the fast mode at $16/$80 per million tokens. TechCrunch also reported the 30% faster figure.
- Output speed: up to 30% faster than Fable 5.1, per Anthropic.
- Leaderboard rank: #1 on Artificial Analysis with Intelligence Index 65, per Artificial Analysis.
- Coding: highest FrontierCode 2.0 score among released models (49.1% at Xhigh effort), per Artificial Analysis and Anthropic's announcement.
Three ways a student can actually use Fable 5.2
The Claude pricing page breaks down the access for each tier.
Pro during launch week
Free until 16 November, up to half of the week's usage credits. After that it runs at API rates on your credits.
Max or Ultra
Fable is included in the plan up to half of the weekly limit. No per-token charges inside that cap.
Pay as you go
Use the API directly at $8 in / $40 out per 1M tokens, or any of the plans' usage-credit systems.
If you are on the Free tier, note: you cannot use Fable 5.2 at all during launch week — you would need to start a Pro trial. This is your only zero-cost window before you decide.
What a student-scale budget can run
A typical prompt-heavy coding session with Fable 5.2 might cost under $1, while a long agent run with many output tokens could cost $5 or more, so a fresher in India can afford it for personal projects if they watch output tokens. The math comes from the announced API rates: input tokens are cheap at $8 per million, but output tokens at $40 per million are where the bill grows.
Say you send a few thousand input tokens to explain a function and get back a few hundred output tokens with a fix. That is a fraction of a cent, per the announced API rates. Now imagine an agent run that generates thousands of output tokens rewriting files across a repo. That is where a single session crosses the dollar mark, and a long one can reach $5. In rupees, using the conversion above, that is roughly ₹400 for a $5 session. A ₹1,000 monthly budget buys you about 2–3 long agent sessions. A $20 budget buys you about 20+. Which one are you? To keep sessions under ₹100, ask for diffs and one function at a time.
The practical rule: keep output tokens low. Ask for diffs, not full files. Ask for the one function, not the whole module.
Here is a quick sanity check on budgets: at ₹84 per dollar, a $5 session is about ₹420. With a ₹1,000 monthly budget, you get about 2 such sessions. A $20 budget gets you roughly 4 of those long sessions. This uses the same conversion as above, so you do not need to do any math yourself.
What you should build with Fable 5.2 this week
Use the free launch week to build a small tool you will actually ship, like a CLI that summarizes your lecture notes or a code review bot for your GitHub repos. Pick something small enough to finish before 16 November, when the free access on Pro ends per the pricing page.
- A CLI that reads a markdown file of lecture notes and prints a bullet-point summary. Input-heavy, low output tokens, so it stays under a dollar for a full semester of notes.
- A code review bot that comments on your own pull requests for style and missing edge cases. It calls Fable on each diff, so keep the review scope to one file per request.
- A CLI that turns your error messages into plain-English explanations. It takes a pasted stack trace, sends it to Fable, and prints a clear cause and fix. This is input-heavy, so it stays under $0.01 per run.
- A script that generates flashcards from your notes. It sends each key topic to Fable to create a question-answer pair, then stores them in a CSV. This uses Fable directly and costs under $0.01 per run.
Each of these fits in the launch week's free window on Pro, and after that they stay affordable because they are input-heavy. The agent-heavy projects are the ones that will eat your budget, so save those for the Max plan or for a weekend when you can watch the token counter.
Open Claude pricing, check your plan, and start a small project with Fable 5.2 before the free launch week ends on 16 November.
